Diffusion Forcing Transformer with History Guidance
Official Interactive Demo for History-guided Video Diffusion
Website
Paper
Code
Pretrained Models- Error or Unexpected Results? Please try again after refreshing the page and ensure you do not click the same button multiple times.
- Performance Issues or No GPU Allocation? Consider running the demo locally (click the dots in the top-right corner). Alternatively, you can subscribe to Hugging Face Pro for an increased GPU quota.
Demo 1: Any Number of Images → Short 2-second Video
Diffusion Forcing Transformer is a flexible model that can generate videos given variable number of context frames.
Demo 2: Single Image → Long 10-second Video
Diffusion Forcing Transformer, with History Guidance, can generate long videos via sliding window rollouts and temporal super-resolution.
Demo 3: Single Image → Extremely Long Video (Navigate with Your Camera Movements!)
History Guidance significantly improves quality and temporal consistency, enabling stable rollouts for extremely long videos.